training RawNet3 with new dataset

clovaai / voxceleb_trainer

In defence of metric learning for speaker recognition

MIT License

1.06k stars 273 forks source link

training RawNet3 with new dataset #158

Closed angelabr closed 1 year ago

angelabr commented 2 years ago

Hi,

When training Rawnet3 with the default config I get this error:

/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:312: operator(): block: [0,0,0], thread: [0,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "index out of bounds" failed.

Changing nClasses to a bigger number solves the problem but it looks like the model doesn't learn as EER gets stuck at 0, could you point me in the right direction to solve this?

Thank you

Angela BR

Jungjee commented 2 years ago

Hi, I think the error might have occurred because you did not set nClasses correctly. Can you confirm if your nClasses is bigger than the actual number of speakers?